Chinese startup Moonshot AI, supported by Alibaba, has launched its new generative AI model, Kimi K2, that aims to rival OpenAI’s ChatGPT and Anthropic’s Claude, particularly in coding tasks.
The model employs a Mixture-of-Experts architecture with 32 billion active parameters within a trillion-parameter framework, delivering strong performance at a lower cost.
Kimi K2 features agentic intelligence, enabling it to autonomously use tools, write and execute code, and handle complex tasks without human input. It is available in two versions: a base model for developers requiring customisation, and an instruct model optimised for general conversational use.
Benchmark tests show Kimi K2 outperforming Claude Opus 4 and OpenAI’s GPT-4.1 in coding and mathematical reasoning, with notable scores on SWE-bench Verified, LiveCodeBench, and the MATH benchmark. This positions it as a powerful tool for enterprise applications.
A key advantage is affordability. Kimi K2 is free via app and browser, with API costs at 14 cents per million input tokens and $2.49 per million output tokens—significantly cheaper than competitors. Moonshot allows unrestricted use but requires branding for commercial products with large user bases or revenues.
While some users have reported occasional inaccuracies, Kimi K2 has been well received, praised for its open-source nature and cost-effectiveness.